For the purpose of identifying attacks on the network systems, a monitoring method is essential. Identification of DDOS attacks is one of the most important concerns now a days in wireless networks. This paper propose...
详细信息
For the purpose of identifying attacks on the network systems, a monitoring method is essential. Identification of DDOS attacks is one of the most important concerns now a days in wireless networks. This paper proposed a security mechanism to identify DDoS attacks using enhanced autoencoder and Deep Neural Network (AE & DNN). In traditional methods, it is very difficult to classify the normal traffic in the network from the malicious one. Therefore some attacks will exist for a long time. So that the network will get affected by various attacks. To overcome these problems autoencoder is combined with Deep Neural Network, where autoencoder is mainly utilized for feature extraction and Deep Neural Network is used for classification. For validation, experiments have been done on the WSN-DS, CICIDS2017, and NSL-KDD standard datasets. From the experiments made, the proposed method outperformed than the CICIDS2017, and NSL-KDD datasets.
Aircraft dynamics modeling is an important part of advanced control law design and flight safety. To address the challenge of longitudinal dynamics modeling using a small amount of flight test data under high-angle- o...
详细信息
Aircraft dynamics modeling is an important part of advanced control law design and flight safety. To address the challenge of longitudinal dynamics modeling using a small amount of flight test data under high-angle- of-attack (AOA) conditions, we propose an effective approach that integrates machine learning with physical mechanisms. First, a low-fidelity aircraft dynamics model based on physical analysis is established. Second, a Residual Transformer (ResTrans) autoencoder is designed to extract temporal and spatial features from flight motion history under high-AOA conditions. These features are then used to compensate for the modeling errors of the low-fidelity model through a deep neural network (DNN)-based fusion module, resulting in a high-fidelity aircraft dynamics model. Moreover, a physics-informed closed-loop multi-step dynamics evolution (PI-CMDE) paradigm is developed for constructing loss functions, ensuring stable and efficient parameter optimization of the high-fidelity model. Finally, a simulation model of a scaled F-16 aircraft is used to generate a small set of high-AOA flight test data for training and testing the high-fidelity model. Experimental results demonstrate that, compared to three representative aircraft dynamics modeling baseline methods, the proposed approach achieves higher modeling accuracy and better generalization performance, highlighting its advanced capabilities.
Highlights What are the main findings? The SSAE-BPNN models can be trained as multiple deep learning models with different predictive performance by using different activation function configurations of the hidden lay...
详细信息
Highlights What are the main findings? The SSAE-BPNN models can be trained as multiple deep learning models with different predictive performance by using different activation function configurations of the hidden layer. The SSAE model can learn the deep fusion features that reflect the change of tool wear from the combination of multi-sensor sensitive features. What is the implication of the main finding? The integrated learning model based on the stacking learning strategy used the SSE-BPNN models with different prediction performance as the primary learners and the Bayesian optimized GBD model as the secondary learner to construct the intergrated deep learning model for tool wear prediction. The proposed model further improved the predictive performance of tool wear prediction model based on deep learning *** What are the main findings? The SSAE-BPNN models can be trained as multiple deep learning models with different predictive performance by using different activation function configurations of the hidden layer. The SSAE model can learn the deep fusion features that reflect the change of tool wear from the combination of multi-sensor sensitive features. What is the implication of the main finding? The integrated learning model based on the stacking learning strategy used the SSE-BPNN models with different prediction performance as the primary learners and the Bayesian optimized GBD model as the secondary learner to construct the intergrated deep learning model for tool wear prediction. The proposed model further improved the predictive performance of tool wear prediction model based on deep learning *** Accurately predicting tool wear in real time is crucial to enhance the tool prognostics and health monitoring system in computerized numerical control (CNC) machining. This paper proposed a novel integrated deep learning model for predicting the wear of milling tools by fusing multi-sensor features. The raw signals of vibration and cut
In this article, improved variational mode decomposition (IVMD) and the online sequential autoencoder multi-kernel broad learning system (OSAEMKBLS) are integrated to recognize epileptic seizure (ES) epochs from both ...
详细信息
In this article, improved variational mode decomposition (IVMD) and the online sequential autoencoder multi-kernel broad learning system (OSAEMKBLS) are integrated to recognize epileptic seizure (ES) epochs from both multichannel and single-channel electroencephalogram (EEG) recordings. The proposed IVMD extracts the optimum number of efficient band-limited intrinsic mode functions (BLIMFs) and the data fidelity factor ( alpha ) using the irregularity index-based Tsallis entropy as a cost function. The designed autoencoder in the proposed OSAEMKBLS architecture is utilized to extract the most elucidative unsupervised signatures from selected informative BLIMFs, chunk by chunk sequentially. These signatures are then fed into the novel supervised kernel trick-based broad learning system for the efficacious recognition of seizure epochs, based on the root mean square error (RMSE) optimal cost function. The efficacy of the proposed IVMD-OSAEMKBLS algorithm is evaluated using benchmark multichannel scalp EEG (sEEG) and single-channel EEG datasets. The proposed method demonstrates higher learning speed, lower computational complexity, better model generalization, and a lower false positive rate per hour (FPR/h) at 0.019. It achieves outstanding recognition accuracy at 99.98% and a short-event recognition time of 42 ms, compared to the IVMD-BLS, IVMD-OSBLS, and IVMD-OSMKBLS methods. Finally, reconfigurable field-programmable gate array (FPGA) hardware is employed to implement the novel IVMD-OSAEMKBLS, developing a computer-aided diagnosis (CAD) system for the automated diagnosis of ES patients. The integrity and expediency of the proposed algorithm endorse secure and admirable accomplishments in seizure detection and recognition.
As the field of cybersecurity has experienced continual changes, up-to-date techniques have become increasingly necessary to analyze and defend against threats. Furthermore, the current methods consistently produce fa...
详细信息
As the field of cybersecurity has experienced continual changes, up-to-date techniques have become increasingly necessary to analyze and defend against threats. Furthermore, the current methods consistently produce false alarms and sometimes completely miss real threats. This paper proposes an approach that integrates secure blockchain technology with data preprocessing, deep learning, and reinforcement learning to enhance threat detection and response capabilities. To secure the exchange of threat intelligence information, a safe blockchain network is used, which comprises Byzantine Fault Tolerance for high data integrity and Zero-Knowledge Proofs for access control. All relevant information is cleaned and standardized prior to analysis. Subsequently, graph convolutional neural networks with autoencoders are trained on large unlabeled sets of threat data to automatically label various types of threats, with the system employing fuzzy logic to rank and score possible threats. Furthermore, we implemented a feedback loop that incorporates reinforcement learning, thereby improving model performance over time according to guidance provided by cybersecurity specialists. The proposed system achieved high accuracy, precision, negative predictive value, and MCC, as well as notably low FPR and FNR values. The results establish that the proposed system is a reliable and effective measure for detecting cyberthreats.
Discovering the anomalies of the steam power system in time can optimize the operating efficiency and avoid major losses. The existing single anomaly detection method is not effective outside its assumptions. Aiming a...
详细信息
Discovering the anomalies of the steam power system in time can optimize the operating efficiency and avoid major losses. The existing single anomaly detection method is not effective outside its assumptions. Aiming at this problem, a new anomaly detection method based on the coupling of thermoeconomics and autoencoder is proposed. This method uses the autoencoder to reconstruct the normal values of the thermoeconomic calculation benchmark and other parameters. Then the endogenous irreversible loss of each component is calculated according to the benchmark. Finally, it is detected together with the reconstruction error of the parameters, and the deviation exceeding the threshold is abnormal. The experimental results show that under the premise of ensuring the precision, the traditional thermoeconomic anomaly detection method, the autoencoder anomaly detection method and the proposed coupling anomaly detection method can detect 58.7 %, 88.9 % and 94 % abnormal samples, respectively. In terms of the accuracy and F1-score, the coupling method is also the highest, reaching 93.9 % and 96.8 % respectively. It is proved that the coupling method is superior to the single thermoeconomic method or the autoencoder method, which is of great significance to ensure the safe and stable operation of the steam power system.
The anomaly detection of lithium-ion batteries for short circuit (SC) faults is crucial to ensure the safety of the energy storage system. Compared to the diagnosis fault of packs, individual cell fault diagnosis lack...
详细信息
The anomaly detection of lithium-ion batteries for short circuit (SC) faults is crucial to ensure the safety of the energy storage system. Compared to the diagnosis fault of packs, individual cell fault diagnosis lacks a reference target, leading to difficulties in effectively detecting whether an abnormality exists. In this paper, a data-driven detection method based on the autoencoder strategy is proposed for early detection of battery faults without pack information. Within, the autoencoder strategy is used to reconstruct the voltage and detect potential faults. Using the generative adversarial network (GAN) framework for model training reduces its overfitting and improves efficiency. In addition, during anomaly detection, due to the lack of battery pack reference, some abnormal voltage changes due to current variations can lead to misdiagnosis. To address this concern, the mixed features input is proposed to reduce the misdiagnosis rate, which incorporates the equivalent circuit model parameters. Experiments demonstrate that the proposed method can accurately detect SC faults, in particular, it can detect some moderate or weak faults within 1.6 h. Compared to other methods, this method has better effectiveness and robustness. The method proposed in this paper is in line with the development trend for big data and opens up new perspectives for the development of energy storage safety technology.
Graph variational autoencoders (VAEs) have been widely used to address the representation problem of graph nodes. However, most existing graph VAEs focus on minimizing reconstruction loss and overlook the uncertainty ...
详细信息
Graph variational autoencoders (VAEs) have been widely used to address the representation problem of graph nodes. However, most existing graph VAEs focus on minimizing reconstruction loss and overlook the uncertainty in the latent distribution and the issue of posterior collapse during training. An Adversarial Regularize Graph Variational autoencoder Based on Encoder Optimization Strategy (MCM-ARVGE) is proposed from the perspective of network structure and loss function. MCM-ARVGE introduces a Multi-dimensional Cloud Generator (MCG) that transforms the traditional encoder, expanding the Gaussian distribution into a Gaussian cloud distribution. Furthermore, MCM-ARVGE employs the idea of adversarial regularization to train the Gaussian cloud distribution, reducing the randomness of the Gaussian cloud distribution. Finally, based on the Gaussian cloud distribution, an effective uncertainty similarity measurement method for cloud distributions is introduced to address the problem of posterior collapse. Experimental results validate the universality and effectiveness of MCM-ARVGE, as it outperforms the baseline model in graph embedding tasks.
Cervical cancer poses a significant global health challenge, necessitating accurate and efficient diagnostic solutions. This study introduces a novel hybrid framework, AutoEffFusionNet, that integrates unsupervised fe...
详细信息
Cervical cancer poses a significant global health challenge, necessitating accurate and efficient diagnostic solutions. This study introduces a novel hybrid framework, AutoEffFusionNet, that integrates unsupervised feature learning through ResNet-based autoencoders with attention mechanisms and supervised learning via transfer learning models. By leveraging the complementary strengths of these approaches, the proposed method achieves enhanced diagnostic accuracy in cervical cancer classification. Genetic algorithms optimize the feature selection process, retaining only the most relevant attributes, thereby addressing feature redundancy and improving computational efficiency. The selected features are then classified using a Support Vector Machine, effectively combining deep learning's feature extraction capabilities with machine learning's robust classification strengths. Additionally, Grad-CAM visualizations are incorporated to highlight critical regions influencing the classification decisions, enhancing interpretability and transparency. The framework was rigorously evaluated on two benchmark datasets, SIPaKMeD, and Mendeley LBC, achieving remarkable accuracies of 99.26% and 100%, respectively. These results demonstrate the effectiveness of the proposed model in addressing key challenges in cervical cancer diagnosis and its potential for deployment in clinical applications.
Recommender methods have been effectively used in both academic and industrial settings. However, the cold start problem with scarce prior information has become the barrier hindering recommender systems from gaining ...
详细信息
Recommender methods have been effectively used in both academic and industrial settings. However, the cold start problem with scarce prior information has become the barrier hindering recommender systems from gaining further improvements. To overcome this issue, this article proposes a novel autoencoder framework referred to as CSRec, which owns the merits of both neural networks and collaborative filtering. Specifically, to search the nearest neighbors for cold start items, CSRec learns item representation and carries out clustering for items via k-means++ method. After that, with the nearest item cluster, CSRec could perform rating prediction for the cold start items through the autoencoder architecture, which could reconstruct the input space directly. Identically, CSRec could also perform cold start recommendations for users through the neural network. In practice, with the autoencoder architecture, CSRec owns powerful capability in computation and representation, which could deeply exploit the inner relationship for items and yield high performance in addressing cold start issues. Moreover, it could enhance the novelty and diversity of cold start recommendations. Experiments on CiaoDVDs and DoubanMovie certificate the superiority of CSRec in addressing the cold start issue, which could yield accurate performance in terms of RMSE, MAE and top-K and outperform other benchmark recommender approaches significantly.
暂无评论